北京信息科技大学学报（自然科学版）2023，Vol.38Issue(6)：33-40,8.DOI:10.16508/j.cnki.11-5866/n.2023.06.005
基于改进YOLOv8s的钢材表面缺陷检测
Steel surface defect detection based on improved YOLOv8s
摘要
Abstract
A steel surface defect detection algorithm based on improved YOLOv8s was proposed to address problems of insufficient feature extraction ability,insufficient feature fusion,slow model convergence speed and poor regression accuracy of YOLOv8s model in the steel surface defect detection task.Firstly,in order to enable the model to focus on more dimensional feature information,part of the C2f modules in the backbone network and neck network of YOLOv8s model were replaced with C2f-Triplet modules.Secondly,in order to enable the model to aggregate contextual information in a larger perceptual region,the nearest neighbor upsampling module in the neck network of the YOLOv8s model was replaced with the content-aware reassembly of features(CARAFE)upsampling operator.Finally,in order to improve the convergence speed and regression accuracy of the model,the CIoU regression loss function of YOLOv8s was replaced with the SIoU loss function.The experimental results show that in the NEU-DET dataset,the improved YOLOv8s steel surface defect detection algorithm achieves a precision improvement of 1.6 percentage points and a mean average precision increase of 2.2 percentage points compared with the original YOLOv8s algorithm.Compared with the current mainstream steel surface defect detection algorithms,the improved YOLOv8s steel surface defect detection algorithm can detect the category and location of steel surface defects more accurately,and the model is relatively small,making it easy to be deployed on mobile devices.关键词
YOLOv8s/钢材表面缺陷检测/C2f-Triplet模块/CARAFE上采样算子/SIoU损失函数Key words
YOLOv8s/steel surface defect detection/C2f-Triplet module/CARAFE upsampling operator/SIoU loss function分类
信息技术与安全科学引用本文复制引用
张文铠,刘佳..基于改进YOLOv8s的钢材表面缺陷检测[J].北京信息科技大学学报（自然科学版）,2023,38(6):33-40,8.基金项目
国家自然科学基金项目(61501464) （61501464）
北京信息科技大学"勤信人才"培育计划(QXTCP C201703) （QXTCP C201703）
